There has been a long-felt need for semiconductor amplifiers for high-power radio-frequency applications. Those skilled in the art know that “radio” frequencies as now understood extends over the entire electromagnetic spectrum, including those frequencies in the “microwave” and “millimeter-wave” regions, and up to light-wave frequencies. Many of these frequencies are very important for commercial purposes, as they include the frequencies at which radar systems, global positioning systems, satellite cellular communications and ordinary terrestrial cellphone systems operate. Development of semiconductor devices to operate with significant power in some of these radio frequencies is extremely important in improving such services.
The first semiconductor amplifiers were made from doped germanium (Ge) materials. These devices tended to be leaky, in that unwanted current flows tended to upset the desired quiescent operating point, and they had relatively low voltage standoff capability, partially as a result of the tendency to leakage. These combined to make the first semiconductor amplifiers little more than playthings, unreliable and with little power capability. It was early on recognized that silicon (Si) semiconductors were theoretically capable of much higher power and exhibited less leakage than germanium semiconductors. Many years of research and development ultimately led to the production of inexpensive Si-based transistors and amplifiers. Silicon transistor amplifiers became more and more capable as better semiconductor architectures were adopted. As an example of improvement in architecture, early transistors such as diffused transistors were three-dimensional, which tended to make heat removal difficult. A major advance was the development of planar transistors, which allowed the chip to be mounted on a heat removing substrate or sink, and reduced the thermal resistance between the active, heat-producing portion of the chip and the heat sink.
More recently, high power radio-frequency semiconductor development has included improved semiconductors, such as gallium arsenide (GaAs), gallium nitride (GaN), and silicon carbide (SiC). Gallium arsenide has been in use for many years because of its inherently greater power capability than silicon, notwithstanding heat transfer capabilities significantly worse than those of silicon. Gallium arsenide devices may be viewed as having reached a limit on their power capability, at about 2 watts per millimeter of gate periphery.
When transistors were first introduced, they were used for the most part as vacuum tubes had been. That is, they were placed in hard-wired sockets connected by wires to other components of a circuit. Eventually, the reliability of transistors became such that the perceived need for sockets was obviated, and the transistor leads were then fused or soldered directly to the circuit, often by means of solder lugs. The inherent frequency limitations of early transistors, coupled with the unwanted stray reactances of the hard-wired circuits, sharply limited the high-frequency frequency response of amplifiers made with this technology. The voltage and current limitations of the transistors also limited them to relatively low-power applications.
Eventually, it was discovered that improved transistors with sufficient high-frequency capability, together with printed-circuit-board techniques which tended to minimize stray reactances, made transistors capable of operation at frequencies approaching about one gigahertz (GHz) (1000 MHz). Unfortunately, such circuits tended to be hand-made, in that the transistors and other components were loaded by hand into printed circuit boards, and differences among the boards so produced led to significant performance differences among presumably identical units.
The introduction of microwave integrated circuits helped to solve the problem of variations among amplifiers used at frequencies above 1 GHz, and to reduce the problem of unwanted stray reactances. The microwave integrated circuit included one or more discrete transistors mounted on a substrate with separate “printed” matching circuits (matching circuits made by integrated-circuit techniques) also mounted on a substrate. Electrical connections were made between the electrodes of the transistor and the matching circuits by the use of wire bonds. Such wire bonds are exceedingly tiny, and when properly applied can provide very repeatable results.
FIG. 1a is a simplified plan view of a prior-art microwave integrated circuit 10, and FIG. 1b is a cross-section thereof. In FIGS. 1a and 1b, an electrically and thermally conductive heat spreader or heat sink 12 provides a planar surface 12us on which other portions are mounted. Microwave integrated circuit 10 includes a planar discrete transistor arrangement 14 made, for example, on a GaAs substrate 16, having a thickness of about 100 micrometers (μm). The thickness of the substrate material is a compromise, in that the substrate should be as thin as possible so as to minimize the cost of GaAs material, but thick enough so that the finished chip is rugged enough to be handled. The active portion of transistor arrangement 14 is located on a thin upper or surface portion designated 16us of substrate 16, about one or two micrometers (μm) thick. This active portion is the so-called “doped” portion of the semiconductor substrate in which control of the transistor electrical conduction takes place. It happens that the control portion 16 is also the location at which unwanted heat generation is at a maximum.
The description herein includes relative placement or orientation words such as “top,” “bottom,” “up,” “down,” “lower,” “upper,” “horizontal,” “vertical,” “above,” “below,” as well as derivative terms such as “horizontally,” “downwardly,” and the like. These and other terms should be understood to refer to the orientation or position then being described or illustrated in the drawing(s), and not to the orientation or position of the actual element(s) being described or illustrated. These terms are used for convenience in description and understanding, and do not require that the apparatus be constructed or operated in the described position or orientation.
The active portion 16 of the transistor arrangement 14 of FIGS. 1a and 1b includes a plurality of individual transistor elements, each including one or more source (S), drain (D), and gate (G) electrodes. The plan view of FIG. 1a shows transistor arrangement 14 as including four separate portions 14a, 14b, 14c, and 14d. While four such portions are shown, there could be more portions or fewer such portions. Each portion 14a, 14b, 14c, and 14d includes a plurality of individual transistors, not separately shown, each of which includes at least one or more source (S), drain (D), and gate (G) electrodes. The source, drain and gate electrodes of the individual transistors must, in operation, be connected to sources and loads. The drain D and gate G electrodes are connected to by way of conductors extending laterally over (or within) the active surface of the transistor arrangement 14. More particularly, the drain D electrodes of each of the plurality of individual transistors of each portion of transistor arrangement 14 are interconnected on the transistor arrangement 14, and led to a bond or bonding pad. Thus, assuming that each transistor portion 14a, 14b, 14c, and 14d contains two individual transistors (but it may contain more or fewer), the two D electrodes of each transistor portion are interconnected by a conductor network. In FIG. 1a, the D electrodes of portion 14a of transistor arrangement 14 are connected together by an interconnecting circuit 18a of a set 18 of interconnecting circuits, the D electrodes of portion 14b of transistor arrangement 14 are connected together by an interconnecting circuit 18b, the D electrodes of portion 14c of transistor arrangement 14 are connected together by an interconnecting circuit 18c, and the D electrodes of portion 14d of transistor arrangement 14 are connected together by an interconnecting circuit 18d. The four interconnecting circuits 18a, 18b, 18c, and 18d communicate with bonding pads 20a, 20b, 20c, and 20d, respectively, of a set 20 of drain bonding pads. Similarly, the gate (G) electrodes of the transistors of the portions of transistor arrangement 14 are interconnected by conductors of a set 22 of interconnecting conductors and led to bonding pads of a set 24 of bonding pads. More particularly, the gate (G) electrodes of portion 14a of transistor arrangement 14 of FIG. 1a are connected together and to a bonding pad 24a by a conductor 22a, the G electrodes of portion 14b of transistor arrangement 14 of FIG. 1a are connected together and to a bonding pad 24b by a conductor 22b, the G electrodes of portion 14c of transistor arrangement 14 of FIG. 1a are connected together and to a bonding pad 24c by a conductor 22c, and the G electrodes of portion 14d of transistor arrangement 14 of FIG. 1a are connected together and to a bonding pad 24d by a conductor 22d. 
A set of planar matching circuit arrangements 30, 32 is mounted on substrate 12 of integrated circuit 10, spaced by small gaps 29, 31 from the transistor arrangement 14. The various bonding pads of sets 20 and 24 are interconnected with corresponding bonding pads of matching circuit arrangements 30 and 32. This connection is made by bond wires which jump over the gaps 29, 31. More particularly, bonding pad 20a is connected by way of a bond wire 34a to a bonding pad 40a on matching circuit arrangement 30, bonding pad 20b is connected by way of a bond wire 34b to a bonding pad 40b on matching circuit arrangement 30, bonding pad 20c is connected by way of a bond wire 34c to a bonding pad 40c on matching circuit arrangement 30, and bonding pad 20d is connected by way of a bond wire 34d to a bonding pad 40d on matching circuit arrangement 30. The matching circuits (not illustrated) associated with matching circuit arrangement 30 make connection to bonding pads 40a, 40b, 40c, and 40d. In addition, the gate bonding pads of set 24 of bonding pads of transistor arrangement 14 are interconnected by bond wires of a set with corresponding bond pads of a set 42 of bond pads of matching circuit 32. Thus, gate bond pad 24a is connected by a bond wire 44a to a bond pad 42a, gate bond pad 24b is connected by a bond wire 44b to a bond pad 42b, gate bond pad 24c is connected by a bond wire 44c to a bond pad 42c, and gate bond pad 24d is connected by a bond wire 44d to a bond pad 42d. 
Design of a transistor suitable for frequencies above about 1 GHz requires that the features such as the electrodes be tiny. This requires that the source, drain and gate electrodes, and their connections, be very close together, which in turn has an impact on the ability to route conductors over the active surface to make electrical connections. For this reason, the source (S) electrodes of the individual transistors of the transistor arrangement 14 are not taken or connected to the side, for lack of surface area in which to make the connections, so the individual transistors are operated in common-source mode by making source connections to the electrically conductive substrate 12 by way of one or more electrically conductive through vias extending vertically through the structure. The source (S) electrodes of the individual transistors of the portions of transistor arrangement 14 of FIGS. 1a and 1b are connected by way of through vias, one of which is illustrated as 26, extending through at least the semiconductor substrate 16 to make contact with the electrically conductive substrate 12. Thus, the substrate 12 becomes part of the electrical circuit.
In an effort to improve reliability of microwave integrated circuits and to avoid the need for manual assembly and manually applied bond wires, efforts were made to incorporate the transistor and the matching circuits onto the same semiconductor substrate, so as to make monolithic microwave integrated circuits (MMICs). Some of the difficulties associated with this type of structure are described in a November 2000 paper entitled Designing at the System Level: What Will Your Power Amplifier Do in the Chip Set?, by Julio Perdomo of Agilent Technologies. In addition to the problems there described, the type of semiconductor used for the common substrate tends to be driven by the requirements of the transistor portion. MMICs are very reliable, and are preferred to discrete circuits for military applications. However, the presence of matching circuits presents a challenge for MMICs. In particular, the matching circuits must have conductors defining, or connecting to, capacitors and inductors, and those conductors when lying above a ground plane (the conductive substrate 12) form planar transmission lines. It is desirable to have narrow conductors which, in the presence of the support substrate, provide characteristic impedance on the order of 50 ohms. Having conductor widths wider than the narrowest possible, however, tends to increase the surface area of the MMIC which is devoted to circuitry, which in turn leads to overall larger size, and to a requirement for more semiconductor material use. On the other hand, very thin conductors can be lossy and difficult to define. Factors that affect the width of a 50-ohm conductor on the surface of a matching circuit are the thickness and the dielectric constant of the semiconductor substrate material. A suitable thickness of GaAs substrate for use with matching circuits is about 100 micrometers (μm), corresponding to about 0.004 inch (4 mils).
The requirements placed on the semiconductor material and its thickness in a MMIC tend to be driven by the needs of the active transistor portion of the MMIC, which tends to relegate the requirements of the matching circuits to secondary importance. This, in turn, tends to make the design of the matching circuits difficult. The best semiconductor substrate for transistor purposes may not be the best from the point of view of the matching circuit, and yet other factors, such as cost, may be significant considerations, as described, for example, in papers entitled Development of GaN Transistor Process for Linear Power Applications, by A. W. Hanson et al. and AlGaN/GaN HFETs fabricated on 100-mm GaN on Silicon (111) substrates, by J. D. Brown et al., all of Nitronex Corporation. For example, Si has a much lower cost than GaAs, but a much lesser ultimate power capability. When cost is an overriding factor, a MMIC must be based on Si. However, the dielectric constant at high frequencies of GaAs is about 13 and its resistivity can be as high as 108 ohm/cm, while those of Si are about 12 and 104 ohm/cm, respectively. The lower resistivity of the Si material translates into more lossy and generally lower impedance matching circuits than would be the case with a GaAs substrate similarly proportioned. The dielectric constant effect can be ameliorated by providing a thinner matching circuit substrate in Si than it would be in GaAs.
Semiconductor materials are known which, due to greater electron band gaps than Si or GaAs, are capable of providing greater power. GaAs is capable of providing about 2 watts per millimeter of gate periphery (2 w/mm). As an example of a material having greater band gap, silicon carbide (SiC) semiconductor materials are better thermal conductors, and have lower losses than Si devices. Silicon carbide devices appear to have an ultimate power capability of about 4 watts per millimeter of gate periphery. Gallium nitride (GaN) has the capability to provide between 5 and 10 w/mm of gate periphery. GaN has a dielectric constant of about 12.5 and resistivity of about 108 ohm-cm. There is substantial interest in development of GaN for monolithic microwave integrated circuits (MMICs) because of its high power capability for the active or transistor portion, and its low loss and modest dielectric constant make it especially suitable for matching circuits. Silicon carbide and gallium nitride devices are only in a developmental state. Among the problems associated with these materials is reliability, which may be as low as 1000 hours mean time before failure (MTBF). These problems may be related to lattice match dislocations between the semiconductor chips and the substrates on which they are mounted. It has been shown that GaN-on-Si can provide good reliability, as described in the abovementioned paper by A. W. Hanson et al., of Nitronex Corporation. The development of monolithic microwave integrated circuits using SIC or GaN materials is proceeding, but it may be many years before reliable, high power MMICs are available using these technologies. In particular, Defense Advanced Research Projects Agency (DARPA) has embarked on a project to develop and improve the MTBF of GaN-on-SiC monolithic microwave integrated circuits to the range of 1E6 hours (one million hours MTBF). It is anticipated that this project may take several years to reach fruition, and it is conceivable that the desired result may never be achieved.
Improved radio-frequency MMICs are desired.